PollenNet: A novel architecture for high precision pollen grain classification through deep learning and explainable...

Area of Science:

• Botany and Environmental Science
• Computer Science and Artificial Intelligence
• Biotechnology and Agricultural Science

Background:

• Accurate pollen grain classification is vital for environmental, agricultural, and allergy research.
• Existing classification methods face challenges due to pollen's complex structures and species diversity.
• Advanced computational approaches are needed to overcome the limitations of traditional pollen identification techniques.

Purpose of the Study:

• To introduce and evaluate PollenNet, a novel deep learning framework for enhanced pollen grain image classification.
• To demonstrate the superior performance of PollenNet compared to existing state-of-the-art methods.
• To improve the accuracy and reliability of pollen identification for ecological and medical applications.

Main Methods:

• Developed PollenNet, a deep learning framework for pollen image classification.
• Implemented a rigorous data preparation pipeline including denoising and image correction.
• Utilized Explainable AI (XAI) for model interpretability and Receiver Operating Characteristic (ROC) curve analysis for performance evaluation.

Main Results:

• PollenNet achieved high performance metrics: 98.45% accuracy, 98.20% precision, 98.40% specificity, 98.30% recall, and 98.25% F1-score.
• The model demonstrated low error rates with Mean Squared Error (MSE) of 0.03 and Mean Absolute Error (MAE) of 0.02.
• ROC analysis confirmed model reliability with a low False Positive Rate (FPR) of 0.016 and False Negative Rate (FNR) of 0.017.

Conclusions:

• PollenNet significantly advances pollen grain classification accuracy and reliability.
• The deep learning framework offers a powerful tool for ecological research and allergy diagnostics.
• This work highlights the potential of AI in addressing complex challenges in biological and environmental sciences.